Open-source AI massive threat to Google and OpenAI, leaked document reveals

Both firms are using large models which aren't suitable for the long run says a senior engineer
Ameya Paleja
Google building in Seattle, USA.
Google building in Seattle, USA.


OpenAI and Google might be fighting to come up on top when it comes to artificial intelligence (AI) technology but neither company is poised to win it, according to a leaked internal document written by a senior Google engineer. The document has been circulating in Silicon Valley for a few months and was recently published by consulting firm Semi-Analysis.

OpenAI shot to fame last year when its conversational AI chatbot ChatGPT made global headlines. Search engine giant Google which has been working in the AI domain for over a decade now and was presumed to be the leader, was caught scrambling to reassert its dominance. Since then, an AI arms race has begun, which each player is looking to dominate.

Google engineer Luke Sernau, however, does not believe either of the companies will ultimately come on top if they continue to proceed ahead down this road. Sernau published a document internally in April, which has been widely circulated privately until it was made public with permission by the consulting firm.

Who will win the AI arms race?

According to Sernau, Google has been looking over its shoulder to see what OpenAI's next move will be, and while these firms have been squabbling, it is open-source AI that is surging ahead. The Google engineer cited large language models running on a smartphone and personal AI that can be finetuned on a laptop in one evening as examples of how technology had evolved.

Open-source AI massive threat to Google and OpenAI, leaked document reveals
Newer AI models can be finetuned on a laptop

Sernau wrote that AI models developed by private organizations still held the edge but would not be able to do so for long. Open-source models were close to achieving the same results at $100 as corporations that were spending billions of dollars. If new iterations were coming up in months at corporations, open-source tech was doing the same in weeks.

Sernau zoomed in on giant models used by Google and OpenAI as the main reason why their progress was being slowed down, in sharp contrast. The open-source community had come across LLaMa from Meta, which was much smaller and easier to work with.

The engineer stressed the need for Google to also shift to smaller models and learn from the open-source community since they were more nimble and could be quickly iterated upon.

In the end, if better AI models would become available for free, clients would not turn to the likes of Google or OpenAI and pay money to use their inferior models.

Add Interesting Engineering to your Google News feed.
Add Interesting Engineering to your Google News feed.
message circleSHOW COMMENT (1)chevron
Job Board